mindlid · Product & AI Workflow
How a user moves through Mindlid v2 — from onboarding to a completed four-session program — and the product architecture principles that support orchestration, multilingual delivery, reliability, and continuity.
Wellness disclaimer: Mindlid is a therapy-inspired, non-clinical emotional wellness and self-help product. It is not a substitute for professional support. If you are in crisis, contact a qualified professional or local emergency services.
Stage 1Onboarding
The user picks a language (English or Turkish), moves through static intro screens, and answers 10 profile questions in a chat-style intake. Answers are collected locally — no AI call per answer — and submitted together at the end.
Architecture principles · onboarding
Profile context
- A structured profile context is created after onboarding
- Input: complete answer set
- Output: reusable profile context for later stages
Design principle
- Profile context is created once and reused across later stages.
- Language handling is kept separate from the core personalization flow.
Stage 2Program creation
The user enters their main challenge (or chooses a program type). Mindlid names the program, writes its description, and builds the four predefined steps. Language handling happens before the core program logic so the same workflow can serve supported languages.
Architecture principles · creation
Language handling
- User input is normalized for the core workflow when needed.
- Localized display is handled separately from program generation.
Program generation
- Creates a program name and description
- Inputs: profile context + main challenge
- Outputs: structured program framing for the user
Program structure
- Builds four predefined session stages
- Program and session display text is localized through the language layer when needed
Stage 3One program, four steps
Every program is the same four-step arc. Each step defines what that part of the program should ask, collect, and generate — and each is completed through one live session. Session data accumulates: every step carries everything learned before it.
Recognize
Focus, trigger, emotions, body sensations, intensity, next small step
Identify
Automatic thought, possible core belief, cognitive distortions, next small step
Test & Expand
Evidence for/against, balanced thought, SMART experiment, next small step
Commit
Action plan, If-Then plan, confidence, ending intensity, next-time reminder
Vocabulary
Step vs session
- A step defines the purpose of that stage; a session is the user's live interaction with it. One program ≈ four sessions.
Framework ≠ workflow
- Program frameworks define the type of guidance; the runtime workflow shown here is the shared delivery pattern.
Stage 4Session runtime — repeated for each session stage
The user talks with Mindlid in short guided exchanges. Each turn, the session orchestrator decides whether to keep exploring or move to controlled completion. Language handling happens invisibly around the core interaction.
Architecture principles · per exchange
Context used
- Current user message
- Profile context + main challenge
- Relevant session context
- Prior cumulative session state
- Current stage configuration
Structured outputs
- User-facing response content
- Stage progress and completion status
- Updated cumulative session state
Operational support
- Language-processing support
- Operational metadata for reliability and observability
Stage 5Session close & what the user sees
When a session completes, its structured state is snapshotted and the app shows the collected session data directly — focus, trigger, emotions with intensities, the plan. This programmatic display is the active user-facing output: there is no AI-generated session summary.
Architecture principles · continuity
State continuity
- Each completed session retains a validated structured snapshot.
- Relevant completed state is carried into the next session to preserve continuity.
Display path
- Free-text values are localized through the language layer when needed.
- Structured values are rendered programmatically for a predictable user-facing result.
Reliability principle
- Key session results are rendered from validated structured state rather than an unconstrained free-form summary.
Stage 6Next session, or program complete
After each completed session the next one activates, carrying the full accumulated state forward. After session 4, the program is complete — the user keeps the whole structured record of what they recognized, identified, tested, and committed to.
Continuity
Cumulative continuity
- Each completed stage adds validated context to the program record.
- Later sessions can build on what was established earlier without exposing implementation-specific state objects.
Public architecture principles
- User-facing results are rendered from structured session state for predictable output.
- Each session uses explicit reliability checks and a controlled completion path.
- Session interactions are intentionally bounded.
- Language handling is separated from the core program logic.
- Validated context carries forward so later sessions can build on earlier work.